ABSTRACT This research applies the correlation between the cumulative float yields of certain density fractions and raw coal ash content, as well as the empirical model in Mayer washability curves, and finally establishes a prediction model of the cumulative float yields of all seven density fractions based on dynamic raw coal ash content. Test results show that the model’s prediction of cumulative yield at each density fraction is quite accurate when the density is higher than 1.4 g/cm3, with the largest forecast error less than 2%. Then, stepwise regression analysis is carried out among the cumulative float ash content of each density fraction and raw coal ash content to establish the model of the real-time cumulative ash content model of each density fraction. The prediction error is less than 1% when the density is higher than 1.4 g/cm3. The floating and sinking components prediction model is relatively precise, which plays a key role in the feedforward compensation model in the automatic control of heavy-medium density in a coal preparation plant.